Anthropic's Fable 5 and Opus 4.8 large language models have undergone a red-team study to assess their adversarial robustness against automated jailbreak attacks1. The evaluation involved generating hundreds of thousands of adversarial attempts across 7,826 harmful intents, spanning a ten-category harm taxonomy. The HackAgent framework was utilized to test the models' defenses, with every apparent success independently reviewed. The study's findings have significant implications for the security of LLMs, as Anthropic's developments are reshaping both capability and risk surfaces. The security implications of these models trail the hype cycle, making it essential to address potential vulnerabilities. As LLMs become increasingly prevalent, understanding their robustness against adversarial attacks is crucial for mitigating potential risks. The study's results matter to practitioners, as they highlight the need for continued evaluation and improvement of LLM security to prevent potential misuse.